Comparative Analysis: Effective Information Retrieval Using Different Learning Approach

International Journal of Research and Scientific Innovation (IJRSI) | Volume IV, Issue VII, July 2017 | ISSN 2321–2705
www.rsisinternational.org Page 90
Comparative Analysis: Effective Information
Retrieval Using Different Learning Approach
Vijay Kumar Mishra
Asst. Professor, Department of Computer Application, Feroze Gandhi Institute of Engg. And Technology, Raebareli, India
Abstract – Information Retrieval is the activity of searching
meaningful information from a collection of information
resources such as Documents, relational databases and the
World Wide Web. Information retrieval system mainly consists
of two phases, storing indexed documents and retrieval of
relevant result. Retrieving information effectively from huge
data storage, it requires Machine Learning for computer
systems. Machine learning has objective to instruct computers to
use data or past experience to solve a given problem. Machine
learning has number of applications, including classifier to be
trained on email messages to learn in order to distinguish
between spam and non-spam messages, systems that analyze past
sales data to predict customer buying behavior, fraud detection
etc. Machine learning can be applied as association analysis
through supervised learning, unsupervised learning and
Reinforcement Learning. The goal of these three learning is to
provide an effective way of information retrieval from data
warehouse to avoid problems such as ambiguity. This study will
compare the effectiveness and impuissance of these learning
approaches.
Keywords – Information Retrieval (IR), Machine Learning (ML),
Supervised Machine Learning (SML) , Reinforcement Learning
(RL).
I. INTRODUCTION
achine Learning ,is a type of Artificial Intelligence
provides computers with the ability to learn to behave
more intelligently rather than just storing and retrieving data
items like a database system and other applications would do.
Machine learning is inspired with a variety of academic fields,
including statistics, biology, computer science and
psychology. The basic function of Machine learning is to tell
computers how to automatically find a good predictor based
on past experiences and this job is done by good classifier.
The process of using a model to forecast unknown values
(output variables), using a number of known values (input
variables) is called Classification. The classification process is
performed on data set D which holds following objects:
1. Set size → A = {A1,A2,…..,A|A|} , where |A| denotes the
number of attributes or the size of the set A.
2. Class label→ C: Target attribute; C = {c1, c2,….,c|C|} ,
where |C| is the number of classes and |C|>=2 .
Here ML provides the classification function which relates the
attribute values in A and classes in C on given data set D. The
most important application of ML is in Data mining. As It is
difficult to find the solution of certain problem when multiple
alternatives are available and it may be possible to make
wrong choice of alternative. So, Machine learning can often
be successfully applied to these problems, improving the
efficiency of systems and the designs of machines . In
machine learning algorithms, every instance of particular
dataset is represented by using the same set of features. The
nature of these features could be continuous, categorical or
binary. If instances are given with known labels (i.e. the
corresponding correct outputs) then the learning scheme is
known as supervised , while in unsupervised learning
approach the instances are unlabeled. Through applying these
unsupervised (clustering) algorithms, researchers are
optimistic to discover unknown, but useful, classes of items
[3]. Another kind of machine learning is reinforcement
learning. Here the training information provided to the
learning system by the environment (i.e. external trainer) is in
the form of a scalar reinforcement signal. The learner is not
told which action has to take, as in most forms of machine
learning, but instead must discover which actions yield the
most reward by trying them[2][1].
II. SUPERVISED LEARNING
Supervised learning is the machine learning approach consists
of:
1. Supervised training data to infer a function.
2. Each training data consists of a set of training
examples
3. Each example is a pair consisting of an input object
and a desired output value.
A supervised learning algorithm analyzes the training data
and produces an inferred function, which is called a classifier
(if the output is discrete, see classification) or a regression
function (if the output is continuous, see regression). The
inferred function should predict the correct output value for
any valid input object. This requires the learning algorithm to
generalize from the training data to unseen situations in a
"reasonable" way.
Supervised learning implies learning a mapping between a set
of input variables X and an output variable Y and applying
this mapping to predict the outputs for unseen data.
Supervised learning is the most important methodology in
machine learning and it also has a central importance in the
M

processing of multimedia data.[10] In face recognition,
supervised learning is applied as learning by examples in
terms of face color, structure etc so that it learn to define a
face after several repetition.
Classification architecture in supervised learning can be
understood by the following figure 1:
Fig. 1 Classification Architecture
A. HOW SUPERVISED ALGORITHMS WORKS
Given a set of training examples of the form
{(x1,y1),…..,(xn,yn)}, a learning algorithm seeks a function,
g:X->Y where X is the input space and Y is the output space.
The function g is an element of some space of possible
functions G , usually called the hypothesis space. It is
sometimes convenient to represent g using a scoring function
f:X*Y->R such that R is defined as returning the value that
gives the highest score: g(x)=arg max f(x,y) . Let F denote the
space of scoring functions.
B. PROBLEMS OF SUPERVISED LEARNING
ALGORITHMS
Human beings have natural ability to learn from past
experiences but this is not the case with computer systems.. In
supervised or Inductive machine learning, our main goal is to
learn a target function that can be used to predict the values of
a class. The process of applying supervised ML to a real-
world problem is described in below figure 2.
Fig. 2 Supervised Learning Model
The first step in this learning is to deal with dataset.To
prepare the dataset in a better way, an appropriate expert
could suggest better selection of features. If concerned expert
is not in reach, then the other approach is “brute-force”, which
means to consider each possible choices available in the hope
that the right (informative, relevant) features can be isolated.
However, a dataset collected by the “brute-force” method is
not directly suitable for induction. Ultimately, in most cases it
contains noise and missing feature values, and therefore
requires significant pre-processing [2]. In the next step, data
preparation and data preprocessing is a key function of
researcher in Supervised Machine Learning (SML).
Researchers have given techniques to deal with missing data
issue. Hodge & Austin [4] have conducted a survey of
contemporary techniques for outlier (noise) detection. Karanjit
& Shuchita[5] have also discussed different outlier detection
methods which are being used in different machine learning.
H. Jair [6] has done comparison on 6 different outlier
detection methods by performing experiment on benchmark
datasets and a synthetic astronomical domain.[1]
III. UNSUPERVISED LEARNING APPROACH
In contrast with supervised learning and reinforcement
learning, unsupervised learning has no pre defined target
outputs or environmental evaluations associated with each
input; rather than it considers only what aspects of the input
schema should be captured in the output. Unsupervised
learning studies how systems can learn to represent particular
input patterns in a way that reﬂects the statistical structure of
the overall collection of input patterns.
The power of unsupervised machine learning is that it can
spot important correlation and connection between data points
that no human would think to look for. This learning approach
of AI can identify the signals, pattern and linkage in data sets.
Unsupervised learning in general has a long and distinguished
history. Some early inﬂuences were Horace Barlow (see
Barlow, 1992), who sought ways of characterising neural
codes, Donald MacKay (1956), who adopted a cybernetic-
theoretic approach, and David Marr (1970), who made an
early unsupervised learning postulate about the goal of
learning in his model of the neocortex. The Hebb rule (Hebb,
1949), which links statistical methods to neurophysiological
experiments on plasticity, has also cast a long shadow. [7]
Unsupervised learning is much common in the brain than
supervised learning. For example there are around 106
photoreceptors in each human eye whose activities are
regularly changing with the world we visualize and which
give all the information about the objects, how they are
presented, what the lighting conditions are, etc.[7]
Example of unsupervised learning is Clustering or cluster
analysis. Clustering is the process of grouping the objects of
same type (in some sense) so that an object of a group is
much similar to another object. Assume the situation where
the inputs are the photoreceptor activities generated by

various images of a guava and an apple. In the space of all
possible activities, these particular inputs form two clusters,
with many fewer degrees of variation than 106
, i.e. Lower
dimension. One natural task for unsupervised learning is to
ﬁnd and characterize these separate, low dimensional clusters.
Unsupervised learning is applicable in face recognition. Since
there is no desired output therefore in this situation,
categorization is done to differentiate correctly between the
face of a horse, human and cat. Two classes of method have
been suggested for unsupervised learning. Density estimation
techniques explicitly build statistical models (such as
BAYESIAN NETWORKS) of how underlying causes could
create the input. Feature extraction techniques try to extract
statistical regularities (or sometimes irregularities) directly
from the inputs.[7]
IV. REINFORCEMENT LEARNING
Reinforcement learning, a machine learning approach is
concerned with how agents ought to take actions in an
environment so as to maximize reward.
The basic reinforcement model consists of:
1. A set of environment states S.
2. A set of actions A.
3. Rules of transitioning between states.
4. Rules that determine the scalar immediate reward of
transition and
5. Rules that describe what the agent observes.
The RL method demonstrates a rather targeted generation of
candidate links between textual requirements artifacts (high
environment through a discrete sequence of steps and actions
over time t, where t = 0, 1, 2, 3, etc. At each step t, the agent
evaluates the state st ∈ S, where S is a set of all possible states.
Based on the state st, the agent selects an action at ∈ A (st),
where A is a set of possible actions available to the agent in
state st. As the result of the action taken at the moment t, the
agent gains reward rt+1, and moves to the state st+1. Figure 3
displays the reinforcement learning process:
Fig. 3 Reinforcement Learning
As shown in Figure 3, both environment model and reward
function are responsible factors to find the optimal solution.
The mapping of the state st into action at is determined by a
policy πt. Since each state st can present a set of possible
actions A(st), the policy πt denotes the probabilities of
selecting one of the possible actions determined by the state st.
The mapping of states to actions is represented as πt(s,a), the
probability of selecting action a =at, when state s=st. The
agent’s goal is to maximize the total rewards acquired in the
long run. The reward the agent collects depends upon the
actions it takes.[8]
Two components make reinforcement learning powerful:
1. The use of samples to optimize performance.
2. The use of function approximation to deal with large
environments.
Reinforcement learning can be used in large environments in
any of the following situations:
1. A model of environment is known but an analytic
solution is not available.
2. Only a solution model of the environment is given.
3. The only way to collect information about the
environment is by interacting with it.
The first two of these problems could be considered planning
problems (since some form of model is available), while the
last one could be considered as a genuine learning problem.
V. COMPARISON
Comparison of all three learning approach can be seen in
tabular form:
TABLE I. Comparison of three learning approach
Factors Supervised
Learning
Unsupervised
Learning
Reinforcement
Learning
Form of
information
provided
Each instance
has a known
label. Input is
provided with
corresponding
correct output.
Instances are
unlabeled, there is
no error or reward
signal to evaluate
a optimal solution.
Information is
provided by
environment in the
form of
reinforcement
signals. Correct
input/output pairs
are never presented.
Application Neural
networks.
Cluster based
retrieval of
images.
Robot control,
elevator scheduling,
telecommunications
.
Drawback Learning from
past
experience is
the
challenging
task for
computers.
It does not
necessarily
provides any
insight into what
the correlation and
connection
between data
points mean.
Model of the
environment must
be known.

VI. CONCLUSION
This paper represents a comparative analysis of three learning
approach for IR i.e. Supervised learning, Unsupervised
learning and Reinforcement learning. Due to scope of this
paper, it is very difficult to discuss the strength and
weaknesses of each approach of learning. The selection of
learning approach in ML is mainly depends on nature of task.
All these three learning approaches have applications in
different fields for effective information retrieval. We cannot
state any approach as the best approach or the worst approach,
as each one is applicable on the basis of how the features of
particular approach are helpful to handle the problems.
REFERENCES
[1]. Iqbal Muhammad and Zhu Yan, “Supervised Machine Learning
Approaches: a Survey”, School of Information Sciences and
Technology, Southwest Jiao tong University, China, pp. 946-947.
[2]. S. B. Kotsiantis, “Supervised Machine Learning: A Review of
Classification Techniques”, Informatics, Vol. 31, No. 3, pp. 249-
268, 2007.
[3]. James Cussens, “Machine Learning”, IEEE Journal of Computing
and Control, Vol. 7, No. 4, pp 164-168, 1996.
[4]. Victoria J. Hodge and Jim Austin, “A Survey of Outlier Detection
Methodologies”, Artificial Intelligence Review, Vol. 22, No. 2,
pp. 85-126, 2004.
[5]. Hugo Jair Escalante, “A Comparison of Outlier Detection
Algorithms for Machine Learning”, CIC-2005 Congreso
Internacional en Computation-IPN, 2005.
[6]. Karanjit Singh and Shuchita Upadhyaya, “Outlier Detection:
Applications and Techniques”, International Journal of Computer
Science Issues, Vol. 9, Issue. 1, No. 3, pp. 307-323, 2012.
[7]. Peter Dayan, “Unsupervised Learning”, MIT, pp. 1-2.
[8]. Hakim Sultanov and Jane Huffman Hayes, “Application of
Reinforcement Learning to Requirements Engineering:
Requirements Tracing, IEEE 2013, pp. 52-53.
[9]. Padraig Cunningham, Mathew Cord and Sarah Jane Delany,
“Supervised learning”, pp.21.

Comparative Analysis: Effective Information Retrieval Using Different Learning Approach

More Related Content

What's hot (20)

Similar to Comparative Analysis: Effective Information Retrieval Using Different Learning Approach (20)

More from RSIS International (20)

Recently uploaded (20)

Comparative Analysis: Effective Information Retrieval Using Different Learning Approach